SnapDoczilla MCP server

Offline docs-as-code for any backend and any frontend, driven by your AI agent. SnapDoczilla keeps a documentation/ folder inside your repo: a static HTML site (MkDocs Material, local Mermaid diagrams, ADRs, API map and component pages with real code and real usage) that anyone opens with a double click and any dev refreshes with one command.

This MCP server gives your agent the deterministic half of that workflow. The agent writes the docs from your code; the server installs, tracks what changed since the last documented commit, writes pages safely, and builds the site.

Install

Needs uv (for uvx), git, and bash (on Windows: Git for Windows). Building the site also needs Python 3 on PATH, and internet the first time (MkDocs and Mermaid are downloaded once).

Claude Code

claude mcp add snapdoczilla -- uvx snapdoczilla-mcp

Claude Desktop, Cursor and other clients that use mcpServers JSON

{
  "mcpServers": {
    "snapdoczilla": {
      "command": "uvx",
      "args": ["snapdoczilla-mcp"]
    }
  }
}

Codex CLI (~/.codex/config.toml)

[mcp_servers.snapdoczilla]
command = "uvx"
args = ["snapdoczilla-mcp"]

This server does not read your source code. Your client does, with its own file tools (Claude Code, Cursor and Codex have them; in Claude Desktop add a filesystem server).

Tools

There is also a prompt, document_project, that starts the whole flow.

Typical use

Ask your agent: "Document this project" (or "update the docs"). It checks status, installs if needed, reads the real code, writes the pages, builds, and marks the areas as synced. It never commits or pushes: you review git diff.

Also available as a plain Agent Skill

The same workflow ships as a SKILL.md for Claude Code, Codex, OpenCode, Gemini CLI and more. See the repository for screenshots, the examples/shop result and the full documentation.

MIT License.